EP Ladder Model

Sweden tabben
Lancer
EWTNWC LAN Bronze
Posts: 593
Joined: Dec 30, 2017
ESO: tabben

Re: EP Ladder Model

  • Quote

Post by tabben »

Yo @Development Team , is it possible to make the search function for the EP/RE ladder non-case sensitive?
Image
User avatar
United States of America SoldieR
Pro Player
Posts: 2270
Joined: Feb 22, 2015
ESO: SoldieR
Location: Chi City

Re: EP Ladder Model

Post by SoldieR »

Ah, is that what it is? I thought it was just non-functional
User avatar
New Zealand zoom
Gendarme
Posts: 9314
Joined: Apr 26, 2015
ESO: Funnu
Location: New_Sweland

Re: EP Ladder Model

Post by zoom »

n0el wrote:With the recent announcement of the Escape New World Championship, its scale and use of the EP ladder to determine the 64 players who will compete for the top 8, the ESOC team would like to open a short discussion regarding the model that the ESOC EP ladder is using and propose a new model.

Currently, the EP ladder uses the Elo rating system. Read more about it here. The Elo system has been used in Age of Empires rating dating back to the MS Gaming Zone days and has traveled with the franchise through time. The consistency of seeing the Elo type rating is great, but since the launch of the original Elo ladders for Age of Empires, improved models have come along such as Glicko. As I have watched the top of the ladder recently, with the advent of 2019 Year of the Turk and TT takeover, I have noticed that the rankings are very stale in terms of points gained and the current model is not reactive well to activity, in fact, considering you start at 1600, one could argue it rewards cherry picking games and being inactive. To fix this, we propose going to the Glicko model. Glicko type models are widely used across esports, in games such as CSGO, many Chess servers and federations, TF2, Microsoft's TrueSkill system, and others). Due to the additional factors in the Glicko system, the ratings are much reactive than Elo models. All factors considering, it should provide a better climate and ultimately more accurate ladder for future use.
What do you mean when you say "reactive to activity" – conducive (i.e non-incentivizing) to it!?
France Kaiserklein
Pro Player
Posts: 10278
Joined: Jun 6, 2015
Location: Paris
GameRanger ID: 5529322

Re: EP Ladder Model

  • Quote

Post by Kaiserklein »

@Development Team I was told that the EP Elo formula is supposed to be the same as the JP Elo one. I'm not sure if that's true, but anyway the formulas are different. The EP one is more punishing; if the Elo gap between players is somewhat big (even just 100 Elo gap) it becomes very hard to gain Elo as the higher ranked player, and very easy to lose some.

The formula for 1v1 JP Elo is: ΔR = 16 + (RL - RW) x (16 / 400) with RL is the loser's current Elo and RW the winner's current Elo
So let's take these games for example:
Sans titre.jpg
Before the second game starts, I have 1940 Elo and dicktator has 1747 Elo. Now assuming we're using the JP formula to calculate my new Elo after that second game: we get ΔR = 16 + (1747 - 1940) x (16 / 400) = 8.28. If I lost the game, I'd drop approx 24 Elo. This means, roughly, that I'd have to go 3-1 vs dicktator so that our Elo stays the same. Seems quite reasonable.
Now using the EP formula: as you can see I gained only 1 Elo, instead of 8. If I lost that game, I'd drop approx 35 Elo. So I'd have to go roughly 35-1 vs dicktator so that our Elo doesn't change. That's crazy.

So I think adopting a formula closer to the JP one is desirable, cause the EP formula just incentivizes point sitting and tends to average people's Elo more. Like how is it ever worth playing someone who has a bit lower Elo if you need to have a 30+ to 1 winrate? No wonder people are just point sitting before tournaments, and no wonder we need to use current JP Elo to seed players in the upcoming tournament, as the EP Elo is too volatile...


Besides, EP Elo also doesn't seem to have a cap. You can lose more than 31 Elo in a game, while on JP Elo 31 is the cap. So again, that punishes good players a lot. I think that should be changed too.
Image
Image
Image
LoOk_tOm wrote:I have something in particular against Kaisar (GERMANY NOOB mercenary LAMME FOREVER) And the other people (noobs) like suck kaiser ... just this ..
User avatar
United States of America musketeer925
Retired Contributor
Donator 01
Posts: 2483
Joined: Mar 28, 2015
ESO: musketeer925

Re: EP Ladder Model

Post by musketeer925 »

Our ladder claims to use this formula (ladder.php):
16.0f + (team2Elo - team1Elo) * 2 / playerAmount * (32.0f / 400.0f)
So it looks like, at least, we're doing what we claim:

16.0 + (1746.746 - 1941.572) * 2 / 2 * (32 / 400) = 0.414

Looks like it gets rounded up to 1. If the game was flipped, we see conqueror would get 31.8 Elo for the win and kaiser lose 31.8 Elo.

I think we doubled the coefficient in an attempt to make EP Elo swing a little faster in attempt to reach a faster equilibrium. I think maybe we didn't adjust it properly -- perhaps also the constant 16.0 at the beginning should also be adjusted to 32.0, like this:
32.0f + (team2Elo - team1Elo) * 2 / playerAmount * (32.0f / 400.0f)
This which would result in:

Kaiser wins: +/-16.414
Conquerer wins: +/-47.8

These numbers seem more reasonable with the "faster equilibrium" goal in mind -- now instead of 35-1, you just have to go 3-1 to stay even.
Spoiler
France Kaiserklein
Pro Player
Posts: 10278
Joined: Jun 6, 2015
Location: Paris
GameRanger ID: 5529322

Re: EP Ladder Model

Post by Kaiserklein »

Okay thanks, so the formula is supposed to be different.
Either way, yeah it needs to be fixed. Instead of an equilibrium, elo is just super volatile and inaccurate. And top players stay stuck at basically the same Elo as tier 2 players, because whenever you rank up it becomes impossible to keep your Elo.
I personally think the JP formula works just fine, but yeah there are probably other ways to improve the EP formula.
Image
Image
Image
LoOk_tOm wrote:I have something in particular against Kaisar (GERMANY NOOB mercenary LAMME FOREVER) And the other people (noobs) like suck kaiser ... just this ..
User avatar
United States of America musketeer925
Retired Contributor
Donator 01
Posts: 2483
Joined: Mar 28, 2015
ESO: musketeer925

Re: EP Ladder Model

Post by musketeer925 »

I'd be happy to make an update to this if I can get a definitive word on what we should change it to. Implementing a max amount would also be easy. In any case I think most people agree the current implementation has some issues.

Not sure really where the decision authority lies on this one, maybe @n0el or @Cometk could weigh in?
User avatar
United States of America musketeer925
Retired Contributor
Donator 01
Posts: 2483
Joined: Mar 28, 2015
ESO: musketeer925

Re: EP Ladder Model

  • Quote

Post by musketeer925 »

I've updated to the JP Elo formula. Let the controversy ensue.
User avatar
Kiribati princeofcarthage
Retired Contributor
Posts: 8861
Joined: Aug 28, 2015
Location: Milky Way!

Re: EP Ladder Model

Post by princeofcarthage »

Please also update relevant documentation in case.
Fine line to something great is a strange change.
User avatar
United States of America HeadKilla
Lancer
The Big Game Hunter - Bronze ElephantDonator Best
Posts: 976
Joined: Aug 5, 2019
ESO: Headkilla
Location: Minnesota

Re: EP Ladder Model

Post by HeadKilla »

I just had a chance to read over this. Really hope the changes recently made correct the situation Kaiser highlighted. Assuming the numbers are accurate, this is/was a pretty flawed ranking system at the top level.
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: EP Ladder Model

Post by Goodspeed »

Damn, that explains why no one has reached 2k yet. Yeah, a difference of 200 elo shouldn't mean you need to win 97% (35-1) to break even.

https://en.wikipedia.org/wiki/Elo_rating_system
The difference in the ratings between two players serves as a predictor of the outcome of a match. Two players with equal ratings who play against each other are expected to score an equal number of wins. A player whose rating is 100 points greater than their opponent's is expected to score 64%; if the difference is 200 points, then the expected score for the stronger player is 76%.
User avatar
Italy Garja
Retired Contributor
Donator 02
Posts: 9729
Joined: Feb 11, 2015
ESO: Garja

Re: EP Ladder Model

Post by Garja »

The formula should have a cap just like the JP formula. However the cap should be lower for the higher ranked player. E.g. if a 1900 plays vs a 1600 and lose he should lose let's say 30 points and the other guy gain 50 than what the 1600 gains. This is to take into account the fact that upsets can very much happen for trivial things in this game and to take into account that you're never gonna play enough games after that to gain the Elo back against that same player.
Then there should be another cap for the opposite situation with sort of low cap for both players (both gain and lose few points).
Finally, there should be a limit of rating differential over which the game is simply null and awards zero elo points either way (this is to prevent point-trading and other upsets such as unintentional net drop etc.)
Elo built this way is sort of going to inflate but it should work fine for a long period of time if coefficients are chose wisely.
Image Image Image
User avatar
Netherlands Goodspeed
Retired Contributor
Posts: 13002
Joined: Feb 27, 2015

Re: EP Ladder Model

Post by Goodspeed »

This is to take into account the fact that upsets can very much happen for trivial things in this game
Do you mean disconnects? Because you shouldn't account for RNG-based upsets in your elo model. If the game has RNG that affects win rates, it should be reflected in the elo ratings. If a 1600 wins a disproportionate amount of games against an 1800 (more than 24%) due to RNG, their elos should grow closer to each other. Otherwise they're not elo ratings, they're just random numbers with no predictive power.
User avatar
No Flag Jaeger
Jaeger
Posts: 4492
Joined: Feb 28, 2015

Re: EP Ladder Model

Post by Jaeger »

musketeer925 wrote:I've updated to the JP Elo formula. Let the controversy ensue.
Ah it seems like the ELO hasn't been recalculated though, I was really curious to see the explosion :(
last time i cryed was because i stood on Lego
User avatar
Kiribati princeofcarthage
Retired Contributor
Posts: 8861
Joined: Aug 28, 2015
Location: Milky Way!

Re: EP Ladder Model

Post by princeofcarthage »

HeadKilla wrote:I just had a chance to read over this. Really hope the changes recently made correct the situation Kaiser highlighted. Assuming the numbers are accurate, this is/was a pretty flawed ranking system at the top level.
Its not correct to say its flawed cuz at some point exact same issue is found in JP ELO as well. The only difference is that JP elo gives higher skill gap than EP.
Fine line to something great is a strange change.
User avatar
Italy Garja
Retired Contributor
Donator 02
Posts: 9729
Joined: Feb 11, 2015
ESO: Garja

Re: EP Ladder Model

Post by Garja »

Goodspeed wrote:
This is to take into account the fact that upsets can very much happen for trivial things in this game
Do you mean disconnects? Because you shouldn't account for RNG-based upsets in your elo model. If the game has RNG that affects win rates, it should be reflected in the elo ratings. If a 1600 wins a disproportionate amount of games against an 1800 (more than 24%) due to RNG, their elos should grow closer to each other. Otherwise they're not elo ratings, they're just random numbers with no predictive power.
Not necessarily RNG. Just arbitrarily dumb things this game has. Anyway, the way I formulated does indeed reduce the gap between two differently ranked players if the lower one wins. Just it does in
a non proportional way, over an arbitrary Elo gap (e.g. 200 elo). Over a certain higher Elo gap (e.g. 400) the game should basically be unrated.

Put it simple, over certain ELo gaps games, players shouldnt gain and lose the same amount of Elo. Those games are not even supposed to happen. They essentially happen because 1) Elo is relatively fresh and a top player may want to play with a player of relatively similar skill but who just started playing again and 2) this game is not very active so sometimes rated games are played across two completely different skill leagues. In either case you don't want the higher ranked player to lose a disproportionate amount of points he will never get back, while instead you dont want to take away anything from the lower ranked player who won a supposedly fair game.
This will inflate Elos over time, and it's prone to some sort of point trading but it should work anyway (keep in mind the non equal win/loss of point just happens beyond a certain Elo gap).
Image Image Image

Who is online

Users browsing this forum: No registered users and 15 guests

Which top 10 players do you wish to see listed?

All-time

Active last two weeks

Active last month

Supremacy

Treaty

Official

ESOC Patch

Treaty Patch

1v1 Elo

2v2 Elo

3v3 Elo

Power Rating

Which streams do you wish to see listed?

Twitch

Age of Empires III

Age of Empires IV